Two surfaces, one database
Natural-language capture and detailed board management remain separate experiences while writing to the same source of truth.
EC Idea Board
A real-time team kanban board with a companion AI skill that converts conversational brain dumps into structured, traceable work.
product · live
Product owner and builder · Updated Aug 2026
- Team operations
- AI intake
- Kanban
- Realtime
01 · Problem
Ideas captured in chat lose structure and accountability, while forcing people to open a project-management tool makes early capture unnecessarily expensive.
02 · System
The web app provides full kanban management while a separate Claude skill parses natural-language input, preserves the source and writes structured cards through an authenticated API.
03 · Outcome
The deployed board supports realtime updates, drag-and-drop workflow, filters, activity history and both manual and conversational intake paths.
Context
The design deliberately keeps AI out of the primary board interface. Conversation is the fast-entry surface; the board remains the authoritative operating surface.
Architecture
- 01Next.js board and authenticated API routes
- 02Supabase card, column and activity data
- 03Realtime subscriptions for shared state
- 04Claude skill for structured creation, update and query operations
- 05Bearer-token boundary for machine access
Decisions & trade-offs
Evidence
- Deployed team board with shared-password access
- Companion skill supports create, update, query and confirmed bulk ingestion
- Original input remains traceable alongside the structured card
What this taught me
- AI intake works best as an additional path into a stable operational model, not as a replacement for that model.
Follow the system
Connected work is resolved from the shared content graph.